Powering sustainability through brand and product entity resolution

Provenance is the global leader in sustainability marketing technology, helping brands share credible, fact-checked environmental and social impact claims at the point of sale. Serving brands including Unilever, GANNI, and Arla, their platform's value depends entirely on data quality: every sustainability claim must be traced to a verified source.

Sustainability data arrives from radically different sources, retailer CSVs, scraped brand sites, certification bodies like Leaping Bunny, each with its own naming conventions. A brand might appear as "Arla Foods" in one source, "Arla Foods UK Ltd" from a certification body, and simply "Arla" on their website. Building manual rules to match them would require constant maintenance across 15,000 brand rows and 200,000 product rows.

Provenance integrated Zingg into their Databricks lakehouse using the Python API. Rather than writing rules from scratch, Zingg learned what constitutes a match from a small set of labelled examples, and suggested the uncertain ones itself.

• Zingg automatically identifies uncertain training pairs and asks engineers to label them, no manual dataset building required
• Schema alignment only, no rule-writing, no punctuation stripping, no preprocessing overhead
• Separate matching pipelines for brands (15K rows) and products (200K rows) at variant level